Виржиниагийн Технологийн Институтийн судлаачид сансрын роботуудад гадаргуу болон түүний доорх чулуулгийн бүтцийг таньж, саад бэрхшээлийг бие даан шийдвэрлэх боломж олгох шинэ технологи бүтээжээ.
Ангараг болон бусад гариг дээрх сансрын хөлөг, роверуудын хувьд гадаргуугийн доорх чулуулгийн гүн болон бүтцийг мэдэх нь нэн чухал юм. Энэ нь роботуудад энергиэ хэмнэж, хөдөлгөх боломжгүй саад бэрхшээлийг тойрч гарах эсвэл зайлуулах эсэхээ шийдвэрлэхэд тусалдаг. Виржиниа Тех-ийн эрдэмтэн Иян Руан болон профессор Эрик Комендера нарын IEEE Sensors Journal сэтгүүлд нийтлүүлсэн уг судалгаа нь энэхүү асуудлыг шийдвэрлэх шинэ аргачлалыг санал болгож байна.
Судлаачид уг аргыг туршихдаа элсэнд янз бүрийн хэмжээтэй 18 бетон блокыг өөр өөр гүнд булж байрлуулжээ. Тэд Лазер Допплерийн Виброметр (LDV) ашиглан чулуулгийн ил хэсэгт багажаар цохилт өгч, чичирхийллийн өгөгдлийг хэмжсэн байна. Олж авсан мэдээллийг Multilayer Perceptron (MLP) буюу мэдрэлийн сүлжээний загварт оруулснаар чулуулгийн гүн болон геометрийг нарийвчлалтай тодорхойлох боломжтой болжээ.
Энэхүү хиймэл оюун ухаанд суурилсан загвар нь 11 төрлийн хэмжигдэхүүн ашиглан чичирхийлэл болон чулуулгийн булсан гүн хоорондын хамаарлыг таньдаг. Судлаачдын тайлбарласнаар, чулуулаг болон хөрсний харилцан үйлчлэлээс шалтгаалан гүнийг тодорхойлох нь төвөгтэй байдаг ч хоёр далд давхарга бүхий авсаархан мэдрэлийн сүлжээ нь найдвартай үр дүн үзүүлжээ.
Уг технологи нь сансрын уудмыг судлахаас гадна дэлхий дээр барилгын талбай болон байгалийн гамшгийн бүсэд аюултай материалыг илрүүлэхэд ашиглах боломжтой юм. Цаашид судлаачид мэдээллийн санг өргөжүүлж, газар доорх агуй болон тогтворгүй чулуулгийн бүтцийг судлах чиглэлээр уг аргыг үргэлжлүүлэн хөгжүүлэхээр төлөвлөж байна.
Дэлгэрэнгүйг эх сурвалжаас харах
Эх сурвалжийг нээх ↓
Knowing a planet’s surface and subsurface structure is vital when planning robotic and crewed missions to explore it. Before landers, rovers, or crew modules reach the surface, astronauts and mission controllers need to know what they will be landing on. For rovers with a limited energy budget, it’s also crucial that they can determine how far rough, rocky terrain extends beneath the surface. That way, they won’t waste time and energy trying to clear obstacles they can’t possibly move.
In short, scientists and astronauts need to know what’s on a planet’s surface (and beneath it) before they attempt to land there and conduct science operations. Thanks to a team from Virginia Polytechnic Institute and State University (Virginia Tech), scientists could soon have a new set of tools that will allow robotic missions to make these determinations autonomously. These AI-powered tools measure vibrations in surface terrain to determine what lies beneath and how deep structures extend.
The study was led by Yiyan Ruan, a Ph.D. candidate at Virginia Tech, and Erik Komendera, an assistant professor of mechanical engineering (and Ruan’s advisor) who once worked as a researcher at NASA’s Langley Research Center. The paper detailing their findings was published in IEEE Sensors Journal. The tools they have developed could have profound applications for both space exploration and life on Earth.
View generated from the digital terrain model and the nadir and color channels of the High Resolution Stereo Camera on ESA’s Mars Express. Credit: ESA/DLR/FU Berlin
As they indicate in their paper, estimating the burial depth of partially embedded rocks is a key capability for energy-limited robotic excavation. When obstacles like rock formations are in the way, mission teams need to know how deep the rock extends so they can decide whether to remove the obstacle or avoid it. Ruan and Komendera’s solution leverages AI to interpret vibrational data, allowing robots to make those decisions autonomously. As Komendera said in a Virginia Tech press release:
For as long as humans have built things on Earth, we have had to deal with large and heavy obstacles that cannot simply be ignored or scooped out of the way without careful planning. It’s easy enough for us to just look at rocks and make some quick, testable judgments. Is this rock resting or partially buried? Can I move this alone or do I need help and tools? But what if robots need to do this? How can robots make those judgments? Now we have an answer.
To test their approach, their team buried 18 concrete blocks of different shapes and sizes in sand at different depths. They then struck the exposed portions of the blocks with an instrumented hammer and remotely measured the vibrations using a device known as a Laser Doppler Vibrometer (LDV). The results were then input into a Multilayer Perceptron (MLP), a type of neural network that learns to identify complex, non-linear patterns from data.
Their MLP mapped 11 measurable features to identify the relationship between impulse excitation, exposed rock geometry, and burial depth. Given the range of factors to consider, the task was rather complex. But by teaching their AI to account for them, Ruan and Komendera created models that could reliably determine the blocks’ depth.
When the future rovers set down on Mars, they could determine rock depth and obstacle clearance autonomously. Credit: NASA/JPL
“Estimating how deep a rock is buried using vibrations can be tricky,” said Ruan. “The feedback is affected by the rock’s geometry and the way the soil holds it. By carefully selecting features from vibration measurements and exposed geometry, we were able to produce reliable data using a compact neural network with just two hidden layers.”
These models could help robotic missions excavate subsurface rocks, caches of water ice, and remove obstacles along their travel routes. In the meantime, the team plans to keep building datasets and testing their approach for applications on Earth, such as analyzing construction sites and natural disasters to locate where potentially hazardous materials are buried. They could also be used to study unstable rock formations and underground caverns, assisting with resource and construction planning, and advancing exploration of Earth’s subsurface.
Further Reading: Virginia Tech, IEEE Sensors Journal
